THE USE OF DEEP LEARNING TO DETERMINE ARRIVAL TIME OF P AND S WAVES FOR GEOTHERMAL MICROSEISMIC DATA

• Main Author: Haris Fahromi, Iqram
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86280
• Institution: Institut Teknologi Bandung

The output of microseismic data processing depends on the quality of P and S wave picking, which is commonly peformed manually by geophysicist, but this process is time-consuming and prone to human subjectivity. New technologies such as deep learning can help reduce subjectivity and improve efficiency in microseismic data processing. The development of deep learning technology in seismology has shown an improvement in seismicity monitoring performance, especially for P and S wave arrival times picking. However, these deep learning models are trained using regional tectonic earthquake data that has different characteristics from microseismic in the geothermal field. The difference in earthquake characteristics will affect the accuracy and performance of the deep learning model. This final project research focuses on training the PhaseNet automatic picker model using microseismic data and evaluating its performance by comparing the predicted arrival time catalog with the manual catalog, and analyzing it using the concept of confusion matrix. The training results produced three automatic picker models, such as: Model 1 which was trained using only microseismic data, and Model 2 and Model 3 which were trained using transfer learning techniques from the pretrained INSTANCE model, with Model 3 adding freezing to the deepest architectural layer. The analysis results of the three models show an improvement in performance compared to the original PhaseNet model, this is shown by the F1 score of the P phase increasing from 0.62 to an average of 0.93, and the F1 score of the S phase from 0.62 to an average of 0.77. These results prove that the use of deep learning models specifically trained with microseismic data can improve their performance in determining the arrival time of P and S waves in geothermal microseismic data